What 50 Articles Taught Me About Spotting AI Writing
Yesterday, I sat down and worked through about 50 articles from my Medium "For You" feed in one session. Worth pausing on what that is: not Medium broadly, not a curated list — it's the algorithm's read of me, based on…
What 50 Articles Taught Me About Spotting AI Writing
By Chris Daily
Yesterday, I sat down and worked through about 50 articles from my Medium "For You" feed in one session. Worth pausing on what that is: not Medium broadly, not a curated list — it's the algorithm's read of me, based on what I've clicked, read, and lingered on. AI, tech, writing, entrepreneurship. The 50 articles were Medium's best guess at what I'd want to read.
I used Claude to go through each one and make the call: human-written, AI-generated, or somewhere in between.
The numbers came out:
- 26 human-written
- 14 AI-generated
- 8 hybrid (5 disclosed, 3 not)
- 2 one-off editing tasks
Roughly half the writing the algorithm served me had AI fingerprints somewhere on it. About a third was straight AI-generated.
Sit with that for a second. This isn't a random Medium sample. This is what Medium's recommendation system, after watching my reading habits closely, decided I would most want to read. And nearly a third of it was machine-written.
A couple of caveats before the patterns. One: this is one feed on one platform on one day. Other reading environments — curated newsletters, paid Substacks, journals — will look different. Two: I'm not naming names. Some of these writers disclosed their AI use, some didn't, and calling specific people out in a public piece isn't the point. The patterns are.
So let's get to those.
The Tells I Kept Seeing
When I lined the AI-generated pieces up against the human ones, the same signals kept showing up. Not foolproof, not magic — but reliable enough to share.
Headline shapes that almost wrote themselves. The AI bucket was full of titles like "Why X Is Falling Apart," "Forget X," "X Will Never Tell You the Truth," "I Asked AI to Do Y," "I Tried 100 Z's." There's a generator-ness to them. They're optimized for the click, not for capturing an actual thought a human had. Human writers occasionally land on hype headlines too — but their pieces tend to deliver something specific underneath. The AI versions deliver structure where substance should be.
The same byline, again and again. Three of the 14 AI-flagged pieces came from one handle. Two more from another. When you see a high-volume account pumping out three "exposé" pieces a week on whatever's trending, that's a content farm pattern, not a writer pattern. Real writers have rhythms. Mills have output.
Smoothness without specificity. The AI pieces almost never had a stuck moment, a weird sentence, or a strange tangent. Everything connected logically and moved forward. Sounds good, right? Except real thinking has friction. Humans pause. Backtrack. Dwell on something for a paragraph longer than they should because it actually bothers them. AI writing optimizes that out. The result reads like a wiki entry pretending to be an essay.
Names with no weight. AI pieces drop names — researchers, CEOs, papers, products — but the references are floating. There's no "I read this paper twice and the second time the methodology bothered me." There's just "according to a recent study by [Lab]…" with the kind of summary you could generate from the press release alone. Humans cite from inside their experience. AI cites from inside its training data.
The closer that closes nothing. AI essays tend to end on a generic uplift — "as we navigate this rapidly evolving landscape…" or "the future will require all of us to…" Human writers either land somewhere specific or refuse to land at all. They don't reach for the universal moral.
What the Human Pieces Looked Like
The human bucket had range. Some were polished named voices with established platforms. Some were raw personal essays with grammar issues. Some were domain experts writing technically about their field. What they shared wasn't quality — it was specificity.
A working journalist writing about local news AI problems wrote from inside the newsroom, not about it.
A scientist writing about SETI brought his own wrestling with the topic, not a summary of what's been said.
A writer telling a hard story about her marriage didn't reach for a frame. She just wrote the thing.
You could feel a person on the other end of those pieces, even when the prose was rough. That was the dividing line — not polish, not credentials, not topic. Presence.
Why This Matters
Here's the part I keep coming back to.
The algorithm picked these articles for me. It watched what I read, scored what I'd probably click, and served up its best guess. And about a third of that best guess was machine-written. The recommendation system isn't filtering for human-vs-AI — it's filtering for engagement, and AI content engages just fine.
We're going to spend the next several years swimming in text shaped by systems like that one. A lot of it will be machine-made and labeled human. A lot of human-made writing will use AI tools without disclosing it. The signal-to-noise ratio of the algorithmic web is going to get worse before it gets better.
Two skills become disproportionately valuable in that environment:
Reading with discernment. Knowing the tells. Trusting your gut when something feels generated. Not because AI writing is always bad — sometimes it's perfectly fine — but because knowing what you're reading lets you weigh it correctly. A summary written by AI is a summary; treat it as one. A thought written by a human is a thought; engage with it as one.
Writing with presence. If you want your writing to stand out in this environment, the move isn't to polish harder. It's to bring more of yourself to the page. The friction. The specificity. The weird detail only you would notice. Those are exactly the things AI smooths away — and exactly what makes a reader trust you.
This is the human-centered AI argument applied to writing. The tools amplify what you bring. If you bring nothing, they produce nothing worth reading. If you bring your real perspective, they help you sharpen it. The piece in the middle — where AI does the bringing and you do the cleanup — is the slop zone, and readers are going to keep getting better at spotting it.
The Quick Cheat Sheet
When you're reading something and want a fast gut check, run through these:
- Does the headline sound like a template?
- Does the byline have a high volume of similar-shaped pieces?
- Does the prose move smoothly with no friction or strange moments?
- Are the references name-dropped without weight?
- Does the ending reach for a generic uplift?
Three or more yeses and you're probably reading AI-generated content. That's not a moral judgment — it's just useful information about what you're consuming.
The goal isn't to sniff out AI for sport. It's to read better, write more honestly, and stay grounded in a moment when grounded is going to mean a lot.
That's the work.
Building your own eye for this
The cheat sheet above is the fast version. Think Critically with AI goes further — judgment and evaluation skills for working with AI output generally, not just spotting who wrote a blog post.